AI Tools & Models HOTFeaturedAll
arXiv cs.AIScore65

Indic-Task Gains and Regressions after Reinforcement Learning

Summary

The researchers apply GRPO reinforcement learning to a 30B-A3B mixture-of-experts model in four settings, tracking which abilities on Indic-language tasks get better and which get worse.

Why this matters

Covers model and fine-tune · Relevant to this topic

Open original ↗

Source: arXiv cs.AI · arxiv.org

Topics

Related items